How JIVIQ AI Answers Questions
The JIVIQ AI answers exclusively from your Knowledge Base and nothing else. It does not browse the open internet, it does not borrow information from other businesses, and it never improvises prices, return policies, or service promises. Every response is audited claim-by-claim before sending. When your documentation does not contain an answer, JIVIQ states so plainly instead of guessing—and routes the question to your human team and the Assistant Gaps queue.
Core Grounding Principles
100% Grounded in Your Docs
The AI answers strictly from your imported website URLs, sitemaps, and uploaded PDF guides. Zero external web browsing.
Automated Claim Checking
Every draft is cross-referenced against your retrieved passages sentence-by-sentence. Unsupported claims are discarded instantly.
Assistant Gaps Learning Loop
Questions the AI could not answer are collected into a dedicated queue in your sidebar for one-click admin training.
The 4-Step Answering Pipeline
When a visitor asks a question in your live chat widget, JIVIQ executes a disciplined, four-stage pipeline to generate a trustworthy answer. Unlike generic AI chatbots that guess plausible-sounding answers, JIVIQ operates on strict factual retrieval.
Searches Your Knowledge Base
The visitor's query is analyzed to determine its core semantic meaning. JIVIQ performs high-density semantic vector retrieval across your indexed website pages (Import your website) and uploaded files (Upload documents). The engine extracts only the top authoritative passages containing exact factual matches.
Writes a Reply Using Only What It Found
The AI drafts a clear, concise conversational reply using exclusively the retrieved excerpts. The model is forbidden from consulting open internet sources, training memory from external companies, or speculative inferences.
Checks the Reply Before Sending
Before any text is delivered to the visitor, an automated verification engine compares every statement and figure in the draft against the retrieved source material. If a statement cannot be directly corroborated by your uploaded documents, the draft is rejected.
Sends the Reply — or Admits It Does Not Know
If the answer passes verification with 100% confidence, it is sent directly to the visitor (in Autopilot) or populated as an editable draft in your inbox (in Copilot). If verification fails or knowledge is missing, the AI sends a polite, honest admission and notifies your human team to step in.
The Answer Check & Hallucination Defense
Generic conversational AI chatbots often suffer from hallucinations—confidently fabricating dates, inventing refund terms, or hallucinating features that your business never offered. JIVIQ eliminates this risk through an automated claim verification check.
Nothing the AI writes is sent unchecked
After drafting a reply, a second verification step compares the draft, claim by claim, against the Knowledge Base sections it retrieved. If any claim is unsupported—a number that isn't in your material, a policy it inferred, or a step it made up—the draft is discarded immediately.
The 2–4 Second Latency
You might notice that JIVIQ AI replies take 2 to 4 seconds to appear rather than instantly streaming out syllable-by-syllable. That deliberate pause is the verification check executing in the background. In customer support, an accurate answer delivered in 3 seconds is infinitely more valuable than an instantaneous hallucination.
Close Is Not Supported
The AI adheres to a strict standard: close is not the same as supported. If your documentation states that your team provides "on-site consultations throughout Greater London", and a visitor asks if you service Oxford, the AI will not infer that you do. It will state that Oxford is not listed in its documentation.
Zero Partial Guesses
If an AI draft answers 3 parts of a question accurately but speculates on a 4th part, the check rejects the unverified component. The AI answers only the verified portions and explicitly clarifies what is missing, preventing customer misunderstandings.
When the AI Doesn't Know (Honest Refusal)
No Knowledge Base covers every possible customer question on day one. When a visitor asks something outside your imported content, JIVIQ handles it with complete honesty and immediate escalation.
"I couldn't find that in the knowledge base, so I won't guess. Ask me anything else, or I can pass this to the team."
No corporate jargon. No invented policies. No deceptive deflections. Just an honest, helpful response that protects your company's integrity.
What Happens Behind the Scenes
When the AI issues a refusal or fails verification, two automatic events take place simultaneously:
1. Real-Time Team Escalation
The conversation is flagged with an "Assistant had no answer" status pill. It immediately appears under the Escalated filter on the Chats page at app.jiviq.com. Your team receives instant push notifications and emails so a human representative can take over smoothly.
2. Automatic Gap Logging
The unanswered question is routed into your workspace's Assistant gaps queue. The system preserves the visitor's exact phrasing so you can review what your customers are asking and provide the answer for future visitors.
The Assistant Gaps Feedback Loop
Instead of losing missed questions or forcing your team to manually audit hundreds of chat transcripts, JIVIQ automatically collects missing knowledge into a streamlined workflow called Assistant gaps.
Automatic Collection
Every inquiry that the AI could not confidently answer from your Knowledge Base is captured into the Assistant Gaps list in your admin sidebar.
Smart Deduplication
If 25 different website visitors ask about your holiday shipping cutoffs, JIVIQ shows one entry with an aggregate counter, keeping your queue clean and prioritized.
One-Click Training
Click Answer, type the correct information, and select Add to knowledge base. The AI learns the answer instantly for all future conversations.
Quick Dismissal
For accidental queries, spam, or topics you intentionally choose not to cover in automated chat, click Dismiss to archive the item cleanly.
The Weekly Continuous Improvement Loop
Spending just 5 minutes each week reviewing your Assistant Gaps queue transforms your AI assistant into an increasingly comprehensive support agent. As you teach it the questions customers actually ask in production, refusal rates drop and automated resolution increases naturally.
Real-Time Visitor Transparency & Disclosures
JIVIQ is built on the principle of complete transparency. The AI never pretends to be a human, never conceals its automated nature, and keeps visitors informed of its progress at every step.
Upfront AI Disclosures
At first contact, every visitor reads an explicit disclosure line: "I'm {assistant name}, {your business}'s AI assistant. Want a person instead? Just ask." Visitors are never tricked into believing they are speaking with a human employee.
Visible [AI] Message Tags
Every automated reply carries an unambiguous [AI] badge next to the assistant's name. Even if you name your assistant after a team member or company mascot, the badge clearly distinguishes automated messages from human staff replies.
Live Status Chips
While processing a reply, the widget displays the exact stage of computation: "Searching the docs…", "Writing a reply…", and "Checking every claim against the docs…". This openness reassures visitors that their question is being fact-checked thoroughly.
Four Absolute Guarantees (What the AI Never Does)
To protect your business legalities, customer trust, and brand reputation, JIVIQ enforces strict structural guardrails that cannot be bypassed.
| Guarantee | How JIVIQ Enforces It | Business Benefit |
|---|---|---|
| check_circle Never answers from outside your docs | Vector retrieval is bounded strictly to your uploaded documents and verified website URLs. | Zero risk of citing conflicting third-party blogs or obsolete industry standards. |
| check_circle Never sends unverified drafts | A secondary claim-checking model validates every assertion against retrieved source text before delivery. | Guaranteed protection against hallucinations, invented fees, and made-up policies. |
| check_circle Never borrows other businesses' data | Multi-tenant workspace isolation guarantees your knowledge is private and accessible only by your assistant. | Complete enterprise privacy and competitive confidentiality. |
| check_circle Never poses as a human | Mandatory AI introduction notices, [AI] badges on message headers, and instant handoff upon request. | Maintains customer trust and complies with global AI transparency regulations. |
Expanding and Maintaining Your Knowledge Base
The quality of JIVIQ AI responses directly mirrors the completeness and accuracy of your Knowledge Base. Follow these best practices to ensure comprehensive support coverage:
Keep Your Website Synced
Whenever you update service offerings, pricing pages, or return terms on your website, visit Refresh Knowledge Base Content to re-crawl your URLs and re-index the latest data.
Upload Internal Manuals
Have onboarding guidelines, warranty PDFs, or technical specifications? Upload them directly using Upload Documents so the AI can reference intricate details without publishing them publicly.
Audit Assistant Gaps Weekly
Schedule a recurring 5-minute review of your Assistant Gaps queue to spot emerging questions and answer them before they turn into repeated escalations.
Frequently Asked Questions
Explore detailed answers regarding JIVIQ's retrieval process, automated claim checks, failsafe escalations, and the Assistant Gaps continuous learning loop.
When a visitor types a message, JIVIQ's AI engine instantly extracts the core semantic intent of the question. It performs high-density semantic vector search across your indexed website pages and uploaded documents to retrieve the most relevant passages. The AI only uses what it successfully finds in those passages—it never browses external search engines or pulls unverified outside data.
JIVIQ uses advanced semantic vector retrieval rather than basic keyword matching. This means the AI understands synonyms, natural phrasing, and user intent. If your document mentions 'business operating hours' and a visitor asks 'When are you guys open?', JIVIQ matches the correct information effortlessly.
The AI retrieves content directly from all active sources in your Knowledge Base. This includes crawled website URLs, XML sitemaps, uploaded PDF files, plain text documents, markdown documentation, and manually answered Q&A entries in your Assistant Gaps queue.
Nothing the JIVIQ AI writes is sent unchecked. After drafting a response, a dedicated secondary verification engine audits the draft claim-by-claim against the retrieved Knowledge Base passages. If a single figure, policy, or promise cannot be substantiated by your source documentation, the draft is immediately discarded and the AI admits it does not know.
Unlike generic chatbots that stream unverified words instantly, JIVIQ deliberately takes 2 to 4 seconds to execute its rigorous 4-step pipeline: searching your docs, drafting the reply, and auditing every single claim against source passages. This slight pause guarantees that your business reputation remains protected from incorrect statements.
The entire draft is rejected. JIVIQ never sends partially unverified replies. If the system cannot substantiate a detail—such as an unlisted fee or inferred refund policy—it defaults to an honest refusal and alerts your human team to step in.
The AI delivers a transparent, courteous refusal: "I couldn't find that in the knowledge base, so I won't guess. Ask me anything else, or I can pass this to the team." It never makes vague deflections, invents speculative numbers, or apologizes repeatedly.
When the AI cannot find an answer, the conversation is marked with an 'Assistant had no answer' banner in your inbox and flagged in the Escalated inbox view. Your team immediately receives real-time email and push notifications so an agent can jump in and provide human assistance.
Yes. From their very first interaction, the widget greeting states: 'Want a person instead? Just ask.' If a visitor asks for a human, team member, or representative, the AI immediately pauses automatic replies and transfers the thread to your team's live queue.
The Assistant Gaps queue is JIVIQ's automated learning hub located in the admin app sidebar at Knowledge Base → Assistant Gaps. It automatically logs every customer question that the AI could not answer with high confidence, giving you a prioritized checklist of knowledge gaps to fill.
In the Assistant Gaps view, find the question card, click the 'Answer' button, type your accurate business answer into the text box, and click 'Add to knowledge base'. The AI indexes the answer immediately and will use it in future visitor chats without requiring a website re-crawl.
JIVIQ automatically groups duplicate inquiries. Instead of cluttering your dashboard with 50 repetitive cards, the question appears as a single entry with an aggregate count (e.g., 'Asked by 14 visitors'). Answering it once resolves the gap for all future visitors.
Every conversation begins with a transparent greeting: 'I'm [Assistant Name], [Business Name]'s AI assistant. Want a person instead? Just ask.' In addition, all AI-generated responses carry a distinct [AI] badge next to the assistant name so customers always know who is speaking.
While the AI is working, the chat widget displays real-time status chips indicating exactly what is happening: 'Searching the docs…', followed by 'Writing a reply…', and 'Checking every claim against the docs…'. Visitors appreciate the visible honesty and verification effort.
Never. JIVIQ strictly isolates your workspace data. The AI cannot search Google, crawl external third-party sites, or reference data from any other business on JIVIQ. Your visitors receive answers generated solely from your vetted company documentation.